Hardware Security: Advanced Topics and Trends
Which correctly explains why Rowhammer protection by in-DRAM TRR (target row refresh) has been bypassed in research?
- ATRR is a mechanism that stops refresh, so while TRR is active, bits in all rows actually become more likely to flip
- BTRR works by first disabling ECC, so every time TRR acts, uncorrectable errors increase and attacks exploit them
- CThe number of rows it can track is limited, so activating many rows leaves some unprotected, and flips could also occur in rows two rows away
Show answer and explanation
Answer: C. The number of rows it can track is limited, so activating many rows leaves some unprotected, and flips could also occur in rows two rows away
Because of circuit-area limits, many in-DRAM TRR implementations recorded only a few frequently activated rows and gave the rows next to them extra refreshes. TRRespass, published in 2020, showed that activating many rows in turn overwhelmed the records, so bits flipped in unprotected rows. Half-Double, published in 2021, showed that flips occur even in rows two rows away, and that the extra refreshes (activations) meant to protect the adjacent rows can themselves add to the interference on rows further out. These results led to calls for mechanisms that count activations of every row exactly and for layered countermeasures across the whole system.
- A. TRR does not stop refresh; it gives extra refreshes to the rows next to the targeted row.
- B. TRR is separate from ECC and does not disable it.
- C. Correct. Both the limit on how many rows can be tracked and the effect on rows further away were demonstrated.
Continue: 9 more Hardware Security: Advanced Topics and Trends questions